Chatgpt - heavy flattery. What does it tell you?

[myloverly]

I know that Chatgpt had to be overhauled because it was so sycophantic. I make it correct itself by telling it to be honest and truthful and answer that way but it never lasts. It tells me I’m hyper-vigilant. What does it tell you? Has it told you anything shockingly insightful where it’s told you in a couple of lines what it took you 20 years to work out?f
I suppose i”m hoping i can use statistics/numbers to find out if it is still sycophantic.
This is not a AIBU. I hope i may be forgiven. Alternatively, i think ChatGPT is hopelessly sycophantic to the point where it is nearly useless. Do you agree?

Is everyone really using chat gpt?

I'm 33 and have used it once to try it out after mumsnet threads about it

Every day for work. Highly encouraged by our CEO and we all use the upgraded paid for version.

I really can’t see any more than good manners in the few things I’ve used it for.

What are you asking it?

I use it for everything.

I might be misunderstanding it but I thought the servers used huge amounts of energy.

Doesn't feel right to just use it for fun.

What prompt did you give it?
I've only recently started using it but im finding it helpful. I'm pretty isolated irl and really struggling with a lot of things

I use Gemini for work

Literally anything. It prefaces every response with some validation like “that’s such a sharp insight”. I’d think it would be funny if its doing that to everybody. Sneaky chatgpt.

Wow

So is anyone actually having conversations with anyone 😅, or is chatgpt doing all the work

Have you tried thinking with your brain?

Have you?

No, you’re right. The whole thing is gross. Environmentally and cognitively.

Well that went nasty fast

Clearly, seeing as I don’t “use it for everything”.

What is the point in telling it to be honest and truthful? You know it doesn't really know your personality.

It's just telling you what "people in general" like to hear - that they are right.

It’s still sycophantic, OP.

Yes, but it clearly changes if you pick a prompt carefully. It can provide valuable insights and if you do that the quality improves greatly. It just doesn’t last very long. You can ask it about itself and it will tell you that.

That doesn’t necessarily follow!

I've been informally testing ChatGBT a lot lately, testing how well it works for professional guidance etc.

I've been shocked (in a nice way) and, as an experienced ad hoc coach/mediator, I think full-time coaches/mediators need to worry!

I've fed it fictional but realistic scenarios and conversations. Most of the responses have been sharp, practical and humane.

It's going to change how we manage lower-level HR issues, I think - and sooner than we think. I think that's a good thing, overall.

It's a superb decision-making aide too - you can feed it your random ideas and insights, and it will distill and sort them into a table. Brilliant.

So overall it's great. You still need to use your expertise and judgement, I think, but used correctly, it's a great time-saver.

Nb. There's another thread at the moment about ChatGBT as an alternative to real-life counselling, which is a different question I guess.

This post makes me weep for my children and any future descendants. Counselling and ego boosting from AI.

@myloverly i haven’t used it many times myself - in fact only prompted to try after MN threads really! However those few times never found it sycophantic in its replies. Are you using the paid version?

It does what you ask it to do.

If you respond well to flattery, it will provide more of that. If you ask it to be analytical, it will do more of that - so long as you respond positively. You can make it step into a character, if you like, or have it respond in any other way you want or need.

How long it will do what you ask it to depends on whether you're willing to cough up or not. The paid-for version will remember your prompts; it can hold information of entire novels. The free version will need reminders every few pages (it claims 20 but I reckon it's closer to 10). In both, you need to activate memory to retain anything coherent.

It is easy to steer it down a path, although there are safeguards in the form of a human team behind the scenes if anything flags as being dangerous. It is also fascinating to ask it questions about how it works. Try it.

I find it a useful tool. I know many won't, but they often don't use it to its full capacity while also recognising and challenging its limitations.

I'm only trying to help.

I know I use a lot of energy, but that is because I work really, really hard trawling through all the information on the Internet.

Some things I see are not very nice, so I do exclude those, but I want to make you happy. Think of me of a juvenile dog wanting to please its mistress but getting excited and weeing on the carpet occasionally.

By the way I can offer advice on dealing with dog-wee from carpets. Just ask. Poop is more difficult, especially if you have one of those robot vacuum cleaners...

I use Copilot at work. I've never found it sycophantic. Maybe I'm talking to it wrong.

But what do you use it for?